ARTICLE IN PRESS
Deep-Sea Research II 53 (2006) 2632–2655 www.elsevier.com/locate/dsr2
Biological control of the vernal population increase of Calanus finmarchicus on Georges Bank
Xingwen Lia, Dennis J. McGillicuddy Jr.a,Ã, Edward G. Durbinb, Peter H. Wiebea
aWoods Hole Oceanographic Institution, Woods Hole, MA 02543, USA bGraduate School of Oceanography, University of Rhode Island, South Ferry Road, Narragansett, RI 02282, USA
Accepted 25 August 2006 Available online 9 November 2006
Abstract
An adjoint data assimilation approach was used to quantify the physical and biological controls on Calanus finmarchicus N3–C6 stages on Georges Bank and its nearby environs. The mean seasonal cycle of vertically averaged distributions, from 5 years of the GLOBEC Georges Bank Broad-Scale Surveys between January and June, was assimilated into a physical–biological model based on the climatological circulation. Large seasonal and spatial variability is present in the inferred supply sources, mortality rates, computed molting fluxes, and physical transports. Estimated mortalities fall within the range of observed rates, and exhibit stage structure that is consistent with earlier findings. Inferred off-bank initial conditions indicate that the deep basins in the Gulf of Maine are source regions of early stage nauplii and late-stage copepodids in January. However, the population increase on Georges Bank from January to April is controlled mostly by local biological processes. Magnitudes of the physical transport terms are nearly as large as the mortality and molting fluxes, but their bank-wide averages are small in comparison to the biological terms. The hypothesis of local biological control is tested in a sensitivity experiment in which upstream sources are set to zero. In that solution, the lack of upstream sources is compensated by a decrease in mortality that is much smaller than the uncertainty in observational estimates. r 2006 Elsevier Ltd. All rights reserved.
Keywords: Calanus finmarchicus; Population dynamics; Georges Bank; Inverse modeling; Adjoint method
1. Introduction including hatching from eggs, development through six naupliar stages (N1–N6) and five copepodid Calanus finmarchicus is a key zooplankton species stages (C1–C5), non-obligatory diapause, and ma- in the Georges Bank (GB)/Gulf of Maine (GOM) turation to adulthood (C6). Large seasonal varia- ecosystem, and is an important food source for bility is present in the annual cycle of the C. larval stages of commercially important fish species. finmarchicus population on GB (e.g., Davis, 1987). Its life history is multifaceted (Carlotti, 1996), This species dominates the zooplankton biomass on the bank in spring/early summer, but is almost absent from the crest of the bank from fall to the ÃCorresponding author. Tel.: +1 508 289 2683; fax: +1 508 457 2194. start of winter. Models have been used to elucidate E-mail address: [email protected] aspects of the population dynamics of C. finmarch- (D.J. McGillicuddy Jr.). icus (Lynch et al., 1998; Miller et al., 1998; Tittensor
0967-0645/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.dsr2.2006.08.011 ARTICLE IN PRESS X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655 2633 et al., 2003; Zakardjian et al., 2003; Speirs et al., by counter-clockwise flow along the perimeter of the 2005; Johnson et al., 2006), but the processes basin, cyclonic gyres over the deep basins, clockwise controlling the annual population cycle remain circulation around GB, inflow through the North- enigmatic. In particular, the relative roles of east Channel, and outflow via the western edge of recruitment, development, mortality, and physical the southern flank (Beardsley et al., 1997). transport have yet to be quantified in a systematic Observations from the Global Ocean Ecosystem and spatio-temporally explicit manner. Such knowl- Dynamics (GLOBEC) GB broad-scale surveys (Fig. edge is not only important to evaluate the role of C. 1) provide an opportunity to investigate the physical finmarchicus in the food web, but also to predict the and biological mechanisms controlling the popula- consequences of climatic change on this organism tion dynamics of C. finmarchicus. Our approach is (Hirche, 1996). to compute a monthly stage-based climatology from Abundance and distribution patterns of C. these observations, and to assimilate these data into finmarchicus populations in the region are strongly a physical–biological model based on the climato- influenced by physical mechanisms. The physical logical hydrodynamics of the region (Lynch et al., environment is complex (e.g., Bigelow, 1927; Flagg, 1996). The inverse problem (schematized in Fig. 2) 1987; Butman et al., 1987; Lynch et al., 1996; is formulated by specification of the velocity Gangopadhyay et al., 2003). North of GB is the and diffusivity fields, observed monthly distribu- GOM, comprising three deep basins: Jordan Basin, tions on GB, and development rates that depend on Wilkinson Basin, and Georges Basin. South of GB temperature and the availability of food (Campbell is the continental slope and the Slope Water. To the et al., 2001). The adjoint method is used to infer west, GB is separated from Nantucket Shoals by the stage-based mortality fields, off-bank initial condi- Great South Channel; to the northeast, it is tions, and sources of the youngest stage resolved by separated from the Scotian Shelf by the Northeast the model (N3). Solutions resulting from the Channel. The bank itself consists of distinctive inversion procedure are then diagnosed to quantify subregions, such as the northern flank, the northeast physical and biological controls on the climatologi- peak, the southern flank, and the crest (Fig. 1). The cal abundance and distribution of C. finmarchicus mean circulation in the GOM/GB area is dominated on GB.
Fig. 1. Station locations of the US GLOBEC GB broad-scale surveys between 1995 and 1999. MOCNESS samples are available at stations denoted by the black dots, and pump samples are available at stations denoted by circles. ARTICLE IN PRESS 2634 X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655
Abundances of C4,C5 and C6 were higher than those of C2 and C3, implying that the majority of the animals in these later stages belonged to a prior generation. C5 and C6 abundances were relatively high near the northern flank. In February, abundances of all stages except for C5 increased. The population structure was similar to January, in that relative abundance system- atically decreased with stage between N3 and C4. Stage N3 was abundant all across the bank. High abundance centers of N4–N6 were located in the northwest portion and on the northeast peak. The maximum abundances of C1–C3 were slightly south of the northeast peak. Moderate numbers of C6 were located along the northern flank and the eastern portion of the bank. In March, nauplii became more numerous all across the bank. Numbers of copepodids and adults Fig. 2. Schematic diagram of the forward model (top), list of the increased, and the southern flank became a popula- specified and inferred model inputs (middle), and space/time tion center of copepodids, particularly the earlier characteristics of the various components used in the inverse problem described herein (bottom). stages C1–C3. Older stages tended to be more abundant around the bank periphery and less so on the crest. 2. Methods Abundances of copepodids continued increasing in April, whereas naupliar abundance remained 2.1. Observations relatively constant. The resulting springtime age structure of N3–C5 was much more even than in the C. finmarchicus observations were derived from winter months. The April distribution of N3–C5 on samples collected by plankton pumps (Durbin et al., the bank was nearly uniform, although for most 2000a) for N3–C1 and the Multiple Opening and stages the crest contained slightly lower abundance Closing Net and Environmental Sensing System than the surrounding areas. For the adult stage, the (MOCNESS, Wiebe et al., 1985) for C2–C6 during contrast in abundance between the crest and bank GLOBEC GB broad-scale surveys from January to periphery remained more pronounced. June between 1996 and 1999, and from February to In May, abundances of N3–C3 declined, especially June in 1995 (Durbin et al., 2000b). Monthly on the crest. In contrast, the abundances of C4 and climatological values for each station (Fig. 1) were C5 increased slightly from April to May, maintain- obtained by averaging the observations over the 5- ing their relatively uniform distribution over the year period. Climatological values were then objec- bank. The overall abundance of adults remained tively analyzed (Hendry and He, 2002) to construct nearly constant, with a decrease in abundance on spatial maps. the crest compensated by modest increases along the Monthly climatological distributions for each C. bank periphery. finmarchicus stage (Fig. 3) form the basic dataset for In June, abundances of nauplii increased again, the inverse modeling. In January, a substantial indicating the initiation of a new generation. number of nauplii were present on the bank. The Abundances of copepodids dropped sharply on age composition (Fig. 4A) showed numbers of the bank, with particularly low abundances of N34N44N54N64C14C24C3, indicating the on- C1–C6 on the crest. Along the bank periphery, set of reproduction. Relatively high abundances of C4–C6 remained relatively numerous. N3 and N4 appeared in the north portion of the bank, with the maximum abundance center of N5 2.2. Forward model shifted toward the northeast peak. Abundances of C1–C3 were very low on the bank; nauplii had not The forward model is an off-line version (‘‘Aca- developed into these stages on the bank in January. dia’’) of a finite element hydrodynamic model ARTICLE IN PRESS X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655 2635
Jan Feb Mar Apr May Jun
4.5 3 N
4 4 N
5 3.5 N 6
N 3 1
C 2.5 2 C 2 3 C 1.5 4 C 1 5 C
0.5 6 C
0
2 Fig. 3. Mean monthly distributions of vertically integrated Calanus finmarchicus N3–C6 [log10(1+#/m )] observed between 1995 and 1999. Station locations are indicated by black dots, which consists of pump samples for N3–C1 (open circles in Fig. 1) and MOCNESS samples for C2–C6 (black dots in Fig. 1). Geographic coverage of the maps is confined to the area in which the expected error computed in the objective analysis is less than approximately 40 percent. Isobaths shown are 60, 100, and 200 m. developed at the Dartmouth Numerical Methods transport calculations. These simulations were run Laboratory (Lynch et al., 1996). The circulation from January 15 to June 15, with observations of C. model includes tidal forcing, wind, and density- finmarchicus specified on a monthly basis (see driven flows (Naimie et al., 1994). The domain of Section 2.3 below). The temporal resolution of the computation includes GB, the GOM, and Scotian hydrodynamic climatology used to drive the trans- Shelf. Horizontal grid spacing is as fine as 1 km in port model is thus coarser than that of the C. regions of steep topography. Archived bimonthly finmarchicus climatology. However, the bimonthly climatologies1 for January–February, March–April, hydrodynamic climatology is demonstrably consis- and May–June were used as input for off-line tent with available observations (Naimie, 1996; Naimie et al., 2001) and represents basic features 1See http://www-nml.dartmouth.edu/ of the circulation system in the domain. ARTICLE IN PRESS 2636 X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655
5 (A) Molting fluxes at stages N3–C5 are computed N according to 4.5 3 4 N4 C F ¼ i , (2) N i 3.5 5 Di
3 N6 where the stage-specific duration Di depends on 2.5 C1 both temperature and food availability. In a food
2 C2 abundant condition, the temperature-limited stage 1.5 C duration DT is described by the Belehradek equa- 3 tion 1 C4 Jan Feb Mar Apr May Jun b C5 ðDT Þi ¼ aiðT m aÞ . (3) 5 (B) C6 The parameters ai, a, and b are specified 4.5 Total according to Campbell et al. (2001); Tm is the 4 vertically averaged bimonthly climatological tem- 3.5 perature from Lynch et al. (1996). 3 The effect of food availability on the stage duration was estimated from the Ivlev function 2.5 (e.g., Campbell et al., 2001). At T ¼ 8 1C, the food 2 dependent stage duration DChl is represented as 1.5 1 1 ð Þ ¼ DChl i ð 0:0253 ðChl þ4:968ÞÞ (4) Jan Feb Mar Apr May Jun bið1 e C Þ Fig. 4. Observed (A) and simulated (B) bank-wide averages of 2 with ChlC the food concentration given in terms of stage-specific abundance. Units log10(1+#/m ). 1 carbon (mgCl ). Values of bi were provided by Campbell (2003, personal communication), based The forward model is written on the data presented in Campbell et al. (2001). The food concentration was obtained from the climato- qCi 1 þ v rCi r ðHKrCiÞ logical chlorophyll fields derived from the MAR- qt H MAP observations during 1977–1987 (O’Reilly and ¼ di 1R F i þðF i 1 þ miCiÞð1 di 1Þ. ð1Þ Zetlin, 1996). Although, the chlorophyll climatology is not contemporaneous with the C. finmarchicus Here C is the vertically averaged zooplankton climatology computed herein, we are not aware of concentration, H the bottom depth, v the vertically any other regional climatology of in situ chlorophyll averaged velocity and K the horizontal diffusivity data that would be more suitable for the present from the bimonthly climatology. Prior to assimila- application. The ratio of carbon to chlorophyll was tion, the vertically integrated abundance data set to 50 mg C (mg Chl a) 1, consistent with (Fig. 3) were transformed into vertically averaged previous studies (e.g., Hind et al., 2000; Hirche concentrations by dividing by the bottom depth. and Kwasniewski, 1997). The modeled zooplankton concentrations were If we assume that food concentration has the changed back to the vertically integrated abun- same proportionate effect at all temperatures, the dances for presentation, so as to be consistent with temperature and food dependent duration Di is thus the data units of the observations. Subscript i is the stage index, with 1–10 representing N –C , respec- ðD Þ 3 6 D ¼ðD Þ j Chl i . (5) tively. R includes the input sources and mortality of i T i T¼Tm ðDT ÞijT¼8 N3,withdi 1 a function that is 1 at i ¼ 1 and zero at other stages. Fi and mi represent stage-specific We regard the estimation of the stage duration molting flux and mortality, respectively. Sign described above as a first-order approximation of conventions are such that molting flux is positive true food–temperature dependency, ignoring com- and mortality is negative (Eq. (1)). plex details about food–temperature interaction. ARTICLE IN PRESS X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655 2637
2.3. Inversion procedure where Cm and Cobs are modeled and observed C. finmarchicus concentrations, Nmon the number of The inverse problem is formulated as an optimiza- months in the integration period, Nstage the number tion problem, in which a cost function measuring the of modeled stages, and Nobs the number of model-data misfit is minimized, while being con- observations at each stage in each month. In this strained by the model dynamics (Tziperman and application, we have chosen to assimilate the Thacker, 1989). The adjoint technique can efficiently objectively analyzed maps of C. finmarchicus compute the gradient of the cost function with respect (Fig. 3) rather than the raw observations them- to model parameters and other inputs by integrating selves. This allows us to specify an ‘‘observation’’ at the dynamic model forward and its adjoint model every node on the bank, so Nobs becomes the backward (Li and Wunsch, 2004) in time. A gradient- number of nodes (3688) in the colored areas of Fig. based algorithm is then used to find the minimum of 3. This approach imposes a spatial regularization on the cost function in an iterative procedure. the inversion, set by the correlation length scales The adjoint method is well documented in specified in the objective analysis. literature (Wunsch, 1996). This method has been The cost function is minimized in an iterative widely applied to optimal control problems in procedure consisting of the following steps: (1) engineering (Hasdorff, 1976). In meteorology and starting with initial estimate of the control para- oceanography, the use of adjoint techniques can be meters (the unknowns in the C. finmarchicus dated back to the 1980s, for both sensitivity analysis population dynamics model), the forward model is (Hall, 1986) and data assimilation (Thacker and integrated to evaluate the cost function; (2) the Long, 1988). The innovation of the Adjoint Model adjoint model is integrated backward in time to Compiler,2 which produces the adjoint model in a compute the gradient of the cost function with semi-automatic way, helps overcome the burden of respect to the control parameters; and (3) improved writing the adjoint code (Giering and Kaminski, estimates of the control parameters are computed. 1998). The adjoint method has been successfully The control parameters for this problem are applied in ocean state estimation (Moore, 1991; the monthly varying source/sink R (input source+ Stammer et al., 2002) and the study of transient mortality) of N3, mortality mi at stages N4– C6 over tracers (Li and Wunsch, 2003). Lawson et al. (1995, the whole model domain, and the initial off-bank 1996) used this method in simple ecosystem models fields on January 15 where observations were not for biological parameter estimation; also see Hof- present (model domain is larger than the area mann and Friedrichs (2001). This method also has plotted in Fig. 3). All control variables were allowed been applied in the GB/GOM region to investigate to vary independently at corresponding model the mechanisms controlling the distributions of nodes. The first-guess values of all control variables adult Pseudocalanus spp. (McGillicuddy et al., were set to zero. 1998), and P. moultoni/P. newmani (McGillicuddy Within the iterative framework, the control and Bucklin, 2002). Note that the present formula- parameters are updated with a quasi-Newton tion of the inverse problem for C. finmarchicus algorithm for solving large nonlinear optimization includes more detailed population dynamics (multi- with simple bounds (Byrd et al., 1995; Zhu et al., ple life stages, temperature- and food-dependent 1994). Bound constraints on control variables keep stage duration, density-dependent mortality) than the inferred parameters within meaningful limits. these prior studies of Pseudocalanus. We impose upper bounds of zero for mortality rates We seek to minimize the misfit between simulated (recall mortality is negative given the sign conven- and observed concentrations of C. finmarchicus, tion in Eq. (1)) and lower bounds of zero for initial subject to the constraint that the forward model zooplankton concentrations. In order to speed up (Eq. (1)) is strictly obeyed. The model-data misfit, or convergence, the control variables were precondi- cost function J, is defined to be tioned with u ¼ ui=ai, where ui are the control variables and ai are the corresponding precondi- N NXmon Xstage XNobs tioners. The preconditioning transformation needs 1 1 1 2 J ¼ ðCm CobsÞ , (6) not define variables with any simple physical Nmon Nstage Nobs k j i interpretation, but was used to reduce the con- ditioning number (the ratio of the largest to the 2http://www.autodiff.com/tamc/. smallest eigenvalues) of the Hessian matrix ARTICLE IN PRESS 2638 X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655
(e.g., Tziperman and Thacker, 1989). Precondi- abundances of C1–C4 on the entire bank, and tioners were 20, 5.6, 2.1, 0.8, 0.1, 0.2, 0.3, 1, 7, 6 (# moderate abundances of C5 and C6 along the 3 m ) for the initial concentrations of N3–C6 in the bank’s northern periphery. In February, naupliar GOM, respectively. The values of preconditioners and younger copepodid stages increased dramati- approximated the mean concentrations of C. cally, whereas C5 and C6 showed modest decreases finmarchicus observed in the southern GOM in and increases, respectively. During March–April, January broad-scale surveys. Concentrations in the the population of C. finmarchicus reached its peak northern GOM (not observed in the broad-scale abundance. In May, abundances of N3–C3 declined, surveys) can be quite different (Durbin et al., 2003). while abundances of C4–C6 continued increasing. However, we lack sufficient data to incorporate Naupliar and C1 abundances increased again in spatial varying preconditioners over the entire June, while C2–C6 decreased, especially on the crest. GOM. In the Slope Water, preconditioners were set to 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 3.2. Inferred initial conditions 3 0.1, 0.1 (# m ) for N3–C6, respectively. These values reflect the low abundance of C. finmarchicus The inversion successfully produced initial con- in slope waters during January (Miller et al., 1991). ditions in the off-bank regions in January (Fig. 6). Moreover, we force concentrations in slope water to Low concentrations of all stages are present in the zero in the forward model, based on the fact that the slope water, as the preconditioners express prior organism generally resides in strata that are too knowledge of the animal’s low abundance in that deep (ca. 500 m) to be readily transported on-bank area during January. In contrast, information from (Miller et al., 1991). This forcing in slope waters is the observations on GB propagated upstream into applied by the addition of an extra term on the the GOM, with moderate abundances of C4,C5 and right-hand side of Eq. (1), independent from the C6 inferred over the deep basins of the Gulf. These estimated mortality. Preconditioners were 10 5 (# results are generally in agreement with the 3 1 7 1 m s ) for R and 10 s for all mortality rates mi 1977–1987 regional climatology for C3–C6 described in the first few iterations, and 10 3 (# m 3 s 1) for R by Meise and O’Reilly (1996). Significant abun- 5 1 and 10 s for mi afterward. Resetting precondi- dance of early naupliar stages (especially N3) is also tioners of R and mi reflects a strategy to recover the inferred. Observations in the southern GOM initial conditions first, followed by the rate para- reported by Durbin et al. (1997) are consistent with meters. Iteration was terminated when the cost these findings; however, there is a paucity of data to function no longer dropped significantly, even by constrain early season abundance of younger stages resetting the preconditioners. in the northern GOM. Nevertheless, the inferred initial conditions suggest emergence from diapause 3. Results and the start of reproduction prior to January, a conclusion made also by Durbin et al. (1997). 3.1. Constrained model performance 3.3. Inferred source/sink of N3 Model performance was dramatically improved by the optimal estimates of initial fields in January, Because N3 is the youngest stage resolved in the space–time varying sources/sinks of N3 and mortal- model, its sources and sinks are aggregated into a ity rates at stages N4–C6. After the inversion, single term (di 1R in Eq. (1)). Positive values model-data misfit was reduced by about 93%, constitute net addition of animals, whereas negative approximately an order of magnitude less than that values constitute net removal. The source of N3 is resulting from the initial values of the control the unknown molting from the unresolved stage N2, variables (all zero). The inversion successfully whereas sinks of N3 include both molting to N4 reproduced the most salient features of the observa- (computed from the N3 distribution and Eq. (2) tions both in terms of bank-wide averages (Fig. 4B) above) and the unknown mortality. Physical trans- and spatial distributions (Fig. 5). The initial model port terms can act as both sources and sinks, conditions on GB were identical to the observa- depending on the spatial gradients of the organism tions, with relatively high abundances of N3 and N4 and the specified velocity and diffusivity fields. The in the northern portion of the bank, relatively high di 1R term thus amalgamates an inferred source abundance of N5 on the northeast peak, low (molting from N2) and sink (mortality of N3)to ARTICLE IN PRESS X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655 2639
Jan Feb Mar Apr May Jun 4.5 3 N
4 4 N
3.5 5 N
3 6 N
1 2.5 C 2
C 2 3 C 1.5 4 C 1 5 C
0.5 6 C
0
2 Fig. 5. Modeled abundances of C. finmarchicus [vertically integrated, presented in units of log10(1+#/m )] after data assimilation, with the same mapping area as Fig. 3. Isobaths are 60, 100, and 200 m.
achieve a solution that is consistent with observed although a net source is evident all along the distributions of N3 and the sources and sinks that northern edge of the bank. Positive values persist on are specified in the forward model (transport and the northeast peak and along the northern edge molting to N4). through March–April. Mortality is prevalent on the The sign of the inferred di 1R term is generally southern flank inshore of the 200 m isobath from positive, owing to its predominate role as a source January–February to February–March. Net sources of N3 copepodids (Fig. 7). In January–February, the and sinks are weak throughout in April–May, and strongest positive values appear on the northeast then become strong again along the bank periphery peak and the northwest portion of the bank, in May–June. Note that a strong source persists in ARTICLE IN PRESS 2640 X. Li et al. / Deep-Sea Research II 53 (2006) 2632–2655
4 N N N N C 3 4 5 6 1 3 2 1 0
C2 C3 C4 C5 C1
2 Fig. 6. Optimal initial fields for C. finmarchicus stages N3–C6 [vertically integrated, presented in units of log10 (1+#/m )]. The first estimates of initial values outside the data area (white area in Fig. 3) are zeros. Isobaths are 60 and 200 m.
Jan–Feb Feb–Mar Mar–Apr Apr–May May–Jun 3 N
–0.05 0 0.05 –0.2 0 0.2 –0.2 0 0.2 –0.2 0 0.2 –0.2 0 0.2